University of Twente Student Theses
Predicting Survival and Recurrence-free Survival in GIST Patients Using Deep Learning
Slettenhaar, J.J. (2024) Predicting Survival and Recurrence-free Survival in GIST Patients Using Deep Learning.
|
PDF
3MB |
| Abstract: | Gastrointestinal stromal tumors (GIST) develop in the gastrointestinal tract. Accurate risk stratification and prognosis are essential. This study investigates improving prediction of survival and recurrence-free survival (RFS) in GIST patients by integrating clinical and CT imaging data using deep learning methods. Traditional risk stratification methods, like the NIH classification, are limited, prompting the need for advanced approaches. Data from the Dutch GIST Consortium were used to develop four models: two based on clinical data (1531 patients) and two on imaging data (159 patients). Both a deep learning survival model and a Cox Proportional Hazard (CPH) model were evaluated using clinical features. Results showed comparable performance for both models in predicting survival and RFS. The deep learning model achieved a C-index of 0.64 for survival, similar to the CPH model. However, interpretability analysis indicated differences in the importance of factors like age, mitotic count, and tumor size. Imaging-based models performed poorly, with a survival model C-index of 0.5 and classification model AUC values around 0.57-0.59. In conclusion, the study shows that deep learning models using clinical features can predict survival and RFS in GIST patients comparably to traditional methods, though further refinement is needed for better risk stratification. Predicting prognosis with imaging data remains challenging, requiring larger datasets and improved models. |
| Item Type: | Essay (Master) |
| Clients: | Erasmus Medical Center, Netherlands |
| Faculty: | TNW: Science and Technology |
| Subject: | 44 medicine, 54 computer science |
| Programme: | Biomedical Engineering MSc (66226) |
| Link to this item: | https://purl.utwente.nl/essays/100194 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page